It's Wednesday, 11 pm.
I'm on my third rewrite of a blog post about sustainability data. The introduction still isn't right. The SEO score is stuck at 64/100. And I've got two more articles to finish before Friday.
My partner walks into my workspace: "You're going to burn out."
She's right.
But here's the thing—I love creating content. I love sharing what I know about Data Science, Data Engineering, and how technology can transform sustainability.
What don't I love? The endless hours of:
Researching facts I already know
Rewriting the same concepts differently
Fighting with SEO tools
Formatting and restructuring
Proofreading for the fourth time
That night, I made a decision:
Either I automate this, or I quit content creation entirely.
Dramatic? Maybe. Necessary? Absolutely.
Here's what content creation actually looks like:
For a single blog post:
45 minutes researching (often stuff I already know)
2 hours writing the first draft
45 minutes rewriting because it "doesn't flow"
1 hour optimising for SEO keywords
30 minutes formatting and checking links
25 minutes final proofread
Total: 5+ hours.
And that's if everything goes smoothly. Most weeks? Add another 2 hours for "unexpected issues."
Multiply that by 3-4 posts per month, and you've got a full-time job.
The cruel irony?
I was hired for my expertise in Sustainability and ESG. But I was spending 80% of my time doing tasks that didn't require my knowledge at all.
I was the bottleneck in my own content creation.
Here's the problem with ChatGPT and generic AI tools:
They know everything about everything. Which means they know nothing deeply about anything.
Your expertise gets diluted. Your voice becomes generic. Your insights sound like everyone else's.
But then I had a different thought:
What if I could train an AI specifically on MY knowledge?
Not the internet's knowledge. Mine.
Everything I've learned about:
Data Science methodologies
Data Engineering workflows
Sustainability technology applications
Every mistake, every breakthrough, every "aha moment"
That's when I started building what I now call my Digital Brain Agent.
I've been using Obsidian for years to capture everything I learn. It's essentially my second brain—every insight, every project, every lesson learned, all interconnected.
I had hundreds of notes. Thousands of connections. Years of expertise, just sitting there.
The challenge: How do I teach an AI to access and use this knowledge the way I do?
I built the entire system using n8n (my favourite workflow automation tool) combined with a RAG (Retrieval-Augmented Generation) system. Here's how it works:
Step 1: Knowledge Capture My Obsidian vault contains everything:
Technical concepts I've mastered
Projects I've completed
Problems I've solved
Frameworks I've developed
Step 2: Vectorisation Using n8n, I created a workflow that:
Monitors my Obsidian vault for new or updated notes
Converts them into vector embeddings (mathematical representations that AI can understand)
Stores them in a searchable RAG database. In my case, I chose Pinecone.
Step 3: Intelligent Retrieval When I need content, the agent:
Understands the topic I'm writing about
Searches through MY knowledge base (not the internet)
Pulls relevant insights from my past work
Uses my actual expertise to generate content
Step 4: Content Generation The AI generates:
Blog drafts that sound like me
SEO-optimised content automatically
Meta descriptions based on best practices
Proper structure with my preferred formatting
Time to build: 15 hours spread over two weekends
Time saved per post: 70%
I'm not going to pretend this was smooth sailing.
After spending 15 hours building the system, I hit "run" expecting magic.
Instead, I got complete nonsense.
The AI was pulling random facts from my notes and mashing them together like a drunk person at a party trying to sound smart.
What went wrong?
I'd vectorised EVERYTHING in my Obsidian vault.
Including:
Half-finished thoughts
Random meeting notes
Personal reminders ("buy milk", "call John")
Draft ideas that made no sense out of context
The AI couldn't distinguish between "Key insight on data pipeline efficiency" and "Remember to buy coffee."
I spent 8 hours troubleshooting before I realised the issue.
The solution was elegantly simple:
I created a tagging system in Obsidian:
✅ #knowledge = Feed to AI
✅ #insight = Priority content for the agent
✅ #draft = Not ready yet
❌ Everything else = Ignore
Then I added a filter in my n8n workflow that only processes notes with the right tags.
Re-ran the agent.
Perfect output.
This taught me the most important lesson about AI automation:
It's not about feeding your AI MORE data. It's about feeding it the RIGHT data.
Let me walk you through my current workflow:
Research topic and gather sources
Create outline
Write first draft
Research SEO keywords
Optimise content for those keywords
Rewrite introduction (twice)
Format and structure
Final proofread
Create meta description
Publish
Time: 6+ hours
Energy: Depleted
Friday evening: Ruined
Input topic into my Digital Brain Agent (2 minutes)
"Write about automated data pipelines for sustainability reporting"
Agent does the heavy lifting (automatic)
Searches my knowledge base for relevant insights
Pulls from my past projects and learnings
Generates a structured draft using MY voice and expertise
Optimises for SEO automatically
Creates meta description and keywords
My review and enhancement (25 minutes)
Check factual accuracy (usually 95%+ accurate)
Add personal anecdotes or recent learnings
Adjust tone if needed (rarely necessary)
Add my unique insights that only I can provide
Publish (3 minutes)
Final formatting check
Schedule or publish immediately
Time: 30 minutes
Energy: Still fresh
Friday evening: Mine again
After two months of using my Digital Brain Agent, here's what changed:
Time Savings:
Per post: From 6 hours to 30 minutes (70% reduction)
Per week: 15+ hours saved
Per month: 60+ hours back in my life
Content Output:
Before: 3-4 posts per week (struggling)
After: 10-12 posts per week (comfortable)
Increase: 3x output
Quality Metrics:
SEO performance: testing
Organic traffic: testing
Average time on page: testing
Engagement rate: testing
Personal Metrics:
Stress level: Massively reduced
Creative energy: Available for strategy
Friday evenings: Actually free
Burnout risk: Eliminated
Here's the crucial difference between my Digital Brain Agent and just using ChatGPT:
ChatGPT:
Trained on the entire internet
Generic insights anyone could write
No understanding of your specific expertise
Same voice as everyone else using it
Can't access your proprietary knowledge
My Digital Brain Agent:
Trained specifically on MY knowledge
Unique insights from my experience
Deep understanding of my specific expertise in Data Science + Data Engineering + Sustainability
Sounds exactly like me (because it learned from me)
Gets smarter every time I learn something new
The best part?
It's not static.
Every time I:
Complete a new project
Learn a new technique
Solve a difficult problem
Have a breakthrough insight
I add it to my Obsidian vault, and the agent automatically absorbs it.
It literally grows with me.
Here's exactly what I use:
Knowledge Management:
Obsidian (for capturing and organising my knowledge)
Structured tagging system (#knowledge, #insight, #draft)
Automation & Orchestration:
n8n (the backbone of the entire workflow)
Custom workflows for monitoring, processing, and triggering
AI & Vectorisation:
RAG system for vector storage and retrieval
Pinecone for vectorisation (you could use alternatives)
Claude API for content generation (trained on my knowledge base)
SEO & Publishing:
Automated keyword analysis
Meta description generation
Automated or semi-automated publishing
Cost: Less than 35€/month for all tools combined
Technical skill required: Intermediate (if you can use n8n, you can build this)
1. Don't vectorise everything. Quality > Quantity. Only feed your agent your best, most structured knowledge.
2. Don't skip the tagging system. You need a way to tell your AI what to learn from and what to ignore.
3. Don't expect perfection immediately. My first outputs were rubbish. It took iteration to get it right.
4. Don't remove yourself from the process. The AI handles the grunt work. You add the insights that make it uniquely yours.
5. Don't forget to keep learning. The agent is only as good as what you feed it. Keep adding new knowledge.
This system works brilliantly if:
✅ You create content regularly (3+ times per week)
✅ You have specific expertise to share
✅ You're tired of spending hours on repetitive tasks
✅ You value quality over generic AI output
✅ You're willing to invest 15-20 hours building the system
This probably isn't for you if:
❌ You only publish occasionally
❌ You prefer doing everything manually
❌ You're not comfortable with automation tools
❌ You don't have a knowledge base to work from
Looking back, here's what I'd change:
1. Start with a smaller knowledge base. I tried to vectorise everything at once. Start with your 20-30 best notes.
2. Build the tagging system FIRST. Don't make my mistake. Tag your notes properly before you start.
3. Test with one type of content. I tried to automate everything. Start with blog posts, then expand.
4. Document your workflow as you build. I had to reverse-engineer my own system to write this article.
5. Plan for iteration. Your first version won't be perfect. Budget time for improvements.
Here's what I believe:
Generic AI is creating a content tsunami. Everyone has access to ChatGPT. Everyone can generate mediocre content instantly.
The result? A race to the bottom.
But here's the opportunity:
The content that will cut through isn't generic. It's deeply personal and expertly informed.
By training AI on YOUR specific knowledge, you get:
The speed of AI
The depth of human expertise
A unique voice that can't be replicated
Content that reflects actual experience
This isn't about replacing human creativity.
It's about cloning your expertise so you can focus on what actually requires your unique brain.
The tedious stuff? Let the agent handle it.
The insights, strategy, and creative direction? That's all you.
I've built this. I use it every day. And it's transformed how I work.
If you're spending hours each week creating content and wondering if there's a better way—there is.
I've created a newsletter where I share:
This isn't theory. It's what I actually do, documented in detail.
Subscribers also get:
My complete n8n workflow templates
The RAG system setup guide
Access to technical Q&A sessions
Early access to new automation strategies
→ Join the newsletter and start building your Digital Brain
One or two emails per week. Only when I've got something genuinely useful to share.
You can keep doing content the old way:
6+ hours per post
Late nights and weekends
Constant stress about deadlines
Burnout always lurking
Or you can build a system that works while you sleep.
I'm not saying it's easy to set up. I spent 15 hours building mine, plus another 8 hours fixing my mistakes.
But that 23-hour investment has given me back 60+ hours every single month.
That's not automation. That's freedom.
What will you do with your extra 15 hours a week?
How accurate is the AI-generated content?
With proper training on quality knowledge, I see 85%+ accuracy. I still review everything, but I'm mostly adding personal touches rather than fixing errors.
Won't this make my content sound robotic?
No—because it's trained on MY writing, not generic internet content. It actually sounds more like me than when I'm tired and forcing content out at 11pm.
Do I need to be technical to build this?
You need intermediate technical skills. If you can use n8n and follow detailed instructions, you can build this. It's not coding, but it's not drag-and-drop either.
What if I don't use Obsidian?
The principles work with any knowledge management system. Notion, Roam, even well-organised Google Docs. The key is having structured, tagged knowledge.
How much does it cost?
My entire stack costs less than £50/month. n8n has a free tier, and the AI API costs are minimal if you're not generating thousands of posts.
How long before I see results?
You'll save time on your very first post. The quality improvements come after a few weeks as you refine the system and add more knowledge.
Can I use this for social media content too?
Absolutely. Once you've built the knowledge base and workflow, you can adapt it for LinkedIn posts, Twitter threads, newsletters—anything.
What's the biggest risk?
Over-relying on automation and losing your unique voice. Always review and add your personal insights. The AI should enhance your expertise, not replace it.
PS: If you're reading this thinking, "I need this in my life," don't wait. Every week you delay is another 15 hours lost to manual content creation.
Subscribe to the newsletter and get my free guide: "Building Your Digital Brain Agent: A Step-by-Step Technical Walkthrough"
The difference between staying overwhelmed and reclaiming your time is one decision.
This decision.